The TL;DR:
Ford's own VP of vehicle hardware engineering admitted the company "mistakenly thought that by just introducing artificial intelligence and ingesting the design requirements we had, that would produce a high-quality product." It didn't, so Ford rehired roughly 350 veteran engineers, many of them retirees, to run weekly design reviews and retrain the AI tools that were supposed to replace them. Klarna ran the same experiment first: it cut about 700 customer service roles for an OpenAI-built assistant, watched customer satisfaction drop and complaints climb, then spent 2025 hiring people back.
Ford and Klarna aren't outliers. They're early examples of a pattern that's now showing up in the data across industries. In the Careerminds survey, 31% of companies said rehiring cost more than the layoffs ever saved, and another 42% said the savings and the rehiring costs roughly canceled out. The math that justified the layoffs mostly didn't survive contact with reality.
The tempting explanation is that the AI wasn't good enough yet. That's mostly wrong. The models performed exactly as advertised on the work they were built for. What broke was something else entirely, and it's worth understanding before your company makes the same bet.
Klarna's CEO Sebastian Siemiatkowski said it plainly after reversing course: "We focused too much on efficiency and cost. The result was lower quality, and that's not sustainable." Ford's engineers weren't rehired to write more code. They were rehired to catch what the AI couldn't see, the design flaw no requirement document anticipated, the call that needed twenty years of context instead of a training set.
AI didn't replace your employees. It replaced the part of their job that was easiest to write down, and kept none of the part that mattered when things got hard.
Researchers call the missing piece tacit knowledge, the judgment that runs a business but was never entered into any system a model can train on: why this customer gets an exception, which design spec is technically fine but will fail in the field, how to read a tense negotiation. As one Forbes analysis put it, AI "isn't built for ambiguity, emotion, or edge cases," and the clearest failures come from businesses that removed human judgment too early. Ford and Klarna both cut the people who carried that judgment before anyone had a plan to preserve it.
Not every company that leaned on AI ended up rehiring. IBM automated 94% of routine HR tasks with an AI agent and reduced its HR headcount by a few hundred, but total employment at IBM went up, not down. CEO Arvind Krishna explained that automating the routine "gives you more investment to put into other areas," so IBM redeployed the savings into programmers and salespeople doing higher-judgment work.
A landmark study by Erik Brynjolfsson and colleagues, tracking 5,179 customer support agents using an AI assistant, found productivity rose 14% on average and 34% for the newest, least-experienced workers, while barely moving top performers. The AI worked by capturing what the best agents already knew and teaching it to everyone else. Customer sentiment and employee retention both improved in that study.
The real differentiator isn't which company had the better model. Ford, Klarna, and IBM likely had access to comparable AI capability. The differentiator is whether the company treated AI as a replacement for people or a tool that makes people better at judgment. IBM kept its experts and handed them the tool. Ford and Klarna handed the tool to nobody and had to hire the experts back.
Tier 1: The Routine Layer
The Concept: This is the high-volume, well-documented middle of any job, the part every case study above shows AI handling well.
The Application: Let AI own it fully, but don't treat "AI can do this" as proof the role is redundant. The routine layer is usually only part of what the person in that role actually does.
Tier 2: The Judgment Layer
The Concept: This is where the expensive mistakes live, the exception, the emotional customer, the design flaw nobody wrote a rule for. It's exactly the layer every rehiring story above failed on.
The Application: Keep humans here on purpose, not by accident because you didn't get around to automating it yet. If a decision's wrong answer is costly, a person should be the one making the final call.
Tier 3: The Redeployment Layer
The Concept: IBM's whole advantage was refusing to let AI-driven savings just disappear as headcount cuts.
The Application: Before you eliminate a role, calculate what the routine automation actually freed up, in hours or in budget, and reinvest it in the judgment layer. A tool like Cowork can help identify which parts of a role are genuinely routine versus which parts only look routine until an exception shows up.
You cut before you captured the knowledge. Ford's AI failed because the engineers who understood the design requirements had already been let go by the time anyone realized the model needed their judgment to work correctly. Once that expertise walks out the door, you can't automate around it, you can only rehire it back at a premium.
The fix: Before any AI-driven reduction, document the judgment calls the role makes, not just the tasks it performs, and confirm the AI can actually handle those calls before the person is gone.
You measured savings, not quality. Klarna's efficiency numbers looked incredible right up until customer satisfaction and complaint volume told a different story. Cost savings are visible immediately, quality erosion shows up months later, after the decision is hard to reverse.
The fix: Track quality and rework metrics for at least one full quarter before calling an AI-driven cut a win, not just the number of tickets closed or headcount removed.
You treated AI as a replacement instead of a tool. This is the difference between IBM's outcome and everyone else's in this piece. Same technology, opposite strategy, opposite result.
The fix: For every role AI touches, ask whether it replaces the person or just the routine part of the job. If it's the routine part, the smart move is redeploying that person, not eliminating them.
The lesson here isn't that AI doesn't work. Ford and Klarna both had access to genuinely capable models. The lesson is that AI raises the floor of an operation, it makes the routine faster and more consistent, but it doesn't raise the ceiling on its own. The ceiling is still set by human judgment, and the companies that cut the humans didn't just lose hands, they lost the ceiling.
This is the same idea Biggest Goal keeps coming back to: implementing AI well isn't a tooling decision, it's a people decision. The companies rehiring in a panic bought the tool and fired the operators. The ones pulling ahead kept the operators and handed them the tool.
If you want to stay ahead of stories like this one, Micah curates the AI News Brief every day. No hype, no fluff, just what's actually changing and what it means for how you run your team.
Subscribe at your.biggestgoal.ai
The TL;DR:
Ford's own VP of vehicle hardware engineering admitted the company "mistakenly thought that by just introducing artificial intelligence and ingesting the design requirements we had, that would produce a high-quality product." It didn't, so Ford rehired roughly 350 veteran engineers, many of them retirees, to run weekly design reviews and retrain the AI tools that were supposed to replace them. Klarna ran the same experiment first: it cut about 700 customer service roles for an OpenAI-built assistant, watched customer satisfaction drop and complaints climb, then spent 2025 hiring people back.
Ford and Klarna aren't outliers. They're early examples of a pattern that's now showing up in the data across industries. In the Careerminds survey, 31% of companies said rehiring cost more than the layoffs ever saved, and another 42% said the savings and the rehiring costs roughly canceled out. The math that justified the layoffs mostly didn't survive contact with reality.
The tempting explanation is that the AI wasn't good enough yet. That's mostly wrong. The models performed exactly as advertised on the work they were built for. What broke was something else entirely, and it's worth understanding before your company makes the same bet.
Klarna's CEO Sebastian Siemiatkowski said it plainly after reversing course: "We focused too much on efficiency and cost. The result was lower quality, and that's not sustainable." Ford's engineers weren't rehired to write more code. They were rehired to catch what the AI couldn't see, the design flaw no requirement document anticipated, the call that needed twenty years of context instead of a training set.
AI didn't replace your employees. It replaced the part of their job that was easiest to write down, and kept none of the part that mattered when things got hard.
Researchers call the missing piece tacit knowledge, the judgment that runs a business but was never entered into any system a model can train on: why this customer gets an exception, which design spec is technically fine but will fail in the field, how to read a tense negotiation. As one Forbes analysis put it, AI "isn't built for ambiguity, emotion, or edge cases," and the clearest failures come from businesses that removed human judgment too early. Ford and Klarna both cut the people who carried that judgment before anyone had a plan to preserve it.
Not every company that leaned on AI ended up rehiring. IBM automated 94% of routine HR tasks with an AI agent and reduced its HR headcount by a few hundred, but total employment at IBM went up, not down. CEO Arvind Krishna explained that automating the routine "gives you more investment to put into other areas," so IBM redeployed the savings into programmers and salespeople doing higher-judgment work.
A landmark study by Erik Brynjolfsson and colleagues, tracking 5,179 customer support agents using an AI assistant, found productivity rose 14% on average and 34% for the newest, least-experienced workers, while barely moving top performers. The AI worked by capturing what the best agents already knew and teaching it to everyone else. Customer sentiment and employee retention both improved in that study.
The real differentiator isn't which company had the better model. Ford, Klarna, and IBM likely had access to comparable AI capability. The differentiator is whether the company treated AI as a replacement for people or a tool that makes people better at judgment. IBM kept its experts and handed them the tool. Ford and Klarna handed the tool to nobody and had to hire the experts back.
Tier 1: The Routine Layer
The Concept: This is the high-volume, well-documented middle of any job, the part every case study above shows AI handling well.
The Application: Let AI own it fully, but don't treat "AI can do this" as proof the role is redundant. The routine layer is usually only part of what the person in that role actually does.
Tier 2: The Judgment Layer
The Concept: This is where the expensive mistakes live, the exception, the emotional customer, the design flaw nobody wrote a rule for. It's exactly the layer every rehiring story above failed on.
The Application: Keep humans here on purpose, not by accident because you didn't get around to automating it yet. If a decision's wrong answer is costly, a person should be the one making the final call.
Tier 3: The Redeployment Layer
The Concept: IBM's whole advantage was refusing to let AI-driven savings just disappear as headcount cuts.
The Application: Before you eliminate a role, calculate what the routine automation actually freed up, in hours or in budget, and reinvest it in the judgment layer. A tool like Cowork can help identify which parts of a role are genuinely routine versus which parts only look routine until an exception shows up.
You cut before you captured the knowledge. Ford's AI failed because the engineers who understood the design requirements had already been let go by the time anyone realized the model needed their judgment to work correctly. Once that expertise walks out the door, you can't automate around it, you can only rehire it back at a premium.
The fix: Before any AI-driven reduction, document the judgment calls the role makes, not just the tasks it performs, and confirm the AI can actually handle those calls before the person is gone.
You measured savings, not quality. Klarna's efficiency numbers looked incredible right up until customer satisfaction and complaint volume told a different story. Cost savings are visible immediately, quality erosion shows up months later, after the decision is hard to reverse.
The fix: Track quality and rework metrics for at least one full quarter before calling an AI-driven cut a win, not just the number of tickets closed or headcount removed.
You treated AI as a replacement instead of a tool. This is the difference between IBM's outcome and everyone else's in this piece. Same technology, opposite strategy, opposite result.
The fix: For every role AI touches, ask whether it replaces the person or just the routine part of the job. If it's the routine part, the smart move is redeploying that person, not eliminating them.
The lesson here isn't that AI doesn't work. Ford and Klarna both had access to genuinely capable models. The lesson is that AI raises the floor of an operation, it makes the routine faster and more consistent, but it doesn't raise the ceiling on its own. The ceiling is still set by human judgment, and the companies that cut the humans didn't just lose hands, they lost the ceiling.
This is the same idea Biggest Goal keeps coming back to: implementing AI well isn't a tooling decision, it's a people decision. The companies rehiring in a panic bought the tool and fired the operators. The ones pulling ahead kept the operators and handed them the tool.
If you want to stay ahead of stories like this one, Micah curates the AI News Brief every day. No hype, no fluff, just what's actually changing and what it means for how you run your team.
Subscribe at your.biggestgoal.ai